kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77
  78 #include <asm/pgtable.h>
  79 #include <asm/pgalloc.h>
  80 #include <asm/uaccess.h>
  81 #include <asm/mmu_context.h>
  82 #include <asm/cacheflush.h>
  83 #include <asm/tlbflush.h>
  84
  85 #include <trace/events/sched.h>
  86
  87 #define CREATE_TRACE_POINTS
  88 #include <trace/events/task.h>
  89
  90 /*
  91  * Protected counters by write_lock_irq(&tasklist_lock)
  92  */
  93 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  94 int nr_threads;                 /* The idle threads do not count.. */
  95
  96 int max_threads;                /* tunable limit on nr_threads */
  97
  98 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  99
 100 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 101
 102 #ifdef CONFIG_PROVE_RCU
 103 int lockdep_tasklist_lock_is_held(void)
 104 {
 105         return lockdep_is_held(&tasklist_lock);
 106 }
 107 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 108 #endif /* #ifdef CONFIG_PROVE_RCU */
 109
 110 int nr_processes(void)
 111 {
 112         int cpu;
 113         int total = 0;
 114
 115         for_each_possible_cpu(cpu)
 116                 total += per_cpu(process_counts, cpu);
 117
 118         return total;
 119 }
 120
 121 void __weak arch_release_task_struct(struct task_struct *tsk)
 122 {
 123 }
 124
 125 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 126 static struct kmem_cache *task_struct_cachep;
 127
 128 static inline struct task_struct *alloc_task_struct_node(int node)
 129 {
 130         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 131 }
 132
 133 static inline void free_task_struct(struct task_struct *tsk)
 134 {
 135         kmem_cache_free(task_struct_cachep, tsk);
 136 }
 137 #endif
 138
 139 void __weak arch_release_thread_info(struct thread_info *ti)
 140 {
 141 }
 142
 143 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 144
 145 /*
 146  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 147  * kmemcache based allocator.
 148  */
 149 # if THREAD_SIZE >= PAGE_SIZE
 150 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 151                                                   int node)
 152 {
 153         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 154                                                   THREAD_SIZE_ORDER);
 155
 156         return page ? page_address(page) : NULL;
 157 }
 158
 159 static inline void free_thread_info(struct thread_info *ti)
 160 {
 161         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 162 }
 163 # else
 164 static struct kmem_cache *thread_info_cache;
 165
 166 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 167                                                   int node)
 168 {
 169         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 170 }
 171
 172 static void free_thread_info(struct thread_info *ti)
 173 {
 174         kmem_cache_free(thread_info_cache, ti);
 175 }
 176
 177 void thread_info_cache_init(void)
 178 {
 179         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 180                                               THREAD_SIZE, 0, NULL);
 181         BUG_ON(thread_info_cache == NULL);
 182 }
 183 # endif
 184 #endif
 185
 186 /* SLAB cache for signal_struct structures (tsk->signal) */
 187 static struct kmem_cache *signal_cachep;
 188
 189 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 190 struct kmem_cache *sighand_cachep;
 191
 192 /* SLAB cache for files_struct structures (tsk->files) */
 193 struct kmem_cache *files_cachep;
 194
 195 /* SLAB cache for fs_struct structures (tsk->fs) */
 196 struct kmem_cache *fs_cachep;
 197
 198 /* SLAB cache for vm_area_struct structures */
 199 struct kmem_cache *vm_area_cachep;
 200
 201 /* SLAB cache for mm_struct structures (tsk->mm) */
 202 static struct kmem_cache *mm_cachep;
 203
 204 static void account_kernel_stack(struct thread_info *ti, int account)
 205 {
 206         struct zone *zone = page_zone(virt_to_page(ti));
 207
 208         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 209 }
 210
 211 void free_task(struct task_struct *tsk)
 212 {
 213         account_kernel_stack(tsk->stack, -1);
 214         arch_release_thread_info(tsk->stack);
 215         free_thread_info(tsk->stack);
 216         rt_mutex_debug_task_free(tsk);
 217         ftrace_graph_exit_task(tsk);
 218         put_seccomp_filter(tsk);
 219         arch_release_task_struct(tsk);
 220         free_task_struct(tsk);
 221 }
 222 EXPORT_SYMBOL(free_task);
 223
 224 static inline void free_signal_struct(struct signal_struct *sig)
 225 {
 226         taskstats_tgid_free(sig);
 227         sched_autogroup_exit(sig);
 228         kmem_cache_free(signal_cachep, sig);
 229 }
 230
 231 static inline void put_signal_struct(struct signal_struct *sig)
 232 {
 233         if (atomic_dec_and_test(&sig->sigcnt))
 234                 free_signal_struct(sig);
 235 }
 236
 237 void __put_task_struct(struct task_struct *tsk)
 238 {
 239         WARN_ON(!tsk->exit_state);
 240         WARN_ON(atomic_read(&tsk->usage));
 241         WARN_ON(tsk == current);
 242
 243         task_numa_free(tsk);
 244         security_task_free(tsk);
 245         exit_creds(tsk);
 246         delayacct_tsk_free(tsk);
 247         put_signal_struct(tsk->signal);
 248
 249         if (!profile_handoff_task(tsk))
 250                 free_task(tsk);
 251 }
 252 EXPORT_SYMBOL_GPL(__put_task_struct);
 253
 254 void __init __weak arch_task_cache_init(void) { }
 255
 256 /*
 257  * set_max_threads
 258  */
 259 static void set_max_threads(void)
 260 {
 261         /*
 262          * The default maximum number of threads is set to a safe
 263          * value: the thread structures can take up at most one
 264          * eighth of the memory.
 265          */
 266         max_threads = totalram_pages / (8 * THREAD_SIZE / PAGE_SIZE);
 267
 268         /*
 269          * we need to allow at least 20 threads to boot a system
 270          */
 271         if (max_threads < 20)
 272                 max_threads = 20;
 273 }
 274
 275 void __init fork_init(void)
 276 {
 277 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 278 #ifndef ARCH_MIN_TASKALIGN
 279 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 280 #endif
 281         /* create a slab on which task_structs can be allocated */
 282         task_struct_cachep =
 283                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 284                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 285 #endif
 286
 287         /* do the arch specific task caches init */
 288         arch_task_cache_init();
 289
 290         set_max_threads();
 291
 292         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 293         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 294         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 295                 init_task.signal->rlim[RLIMIT_NPROC];
 296 }
 297
 298 int __weak arch_dup_task_struct(struct task_struct *dst,
 299                                                struct task_struct *src)
 300 {
 301         *dst = *src;
 302         return 0;
 303 }
 304
 305 void set_task_stack_end_magic(struct task_struct *tsk)
 306 {
 307         unsigned long *stackend;
 308
 309         stackend = end_of_stack(tsk);
 310         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 311 }
 312
 313 static struct task_struct *dup_task_struct(struct task_struct *orig)
 314 {
 315         struct task_struct *tsk;
 316         struct thread_info *ti;
 317         int node = tsk_fork_get_node(orig);
 318         int err;
 319
 320         tsk = alloc_task_struct_node(node);
 321         if (!tsk)
 322                 return NULL;
 323
 324         ti = alloc_thread_info_node(tsk, node);
 325         if (!ti)
 326                 goto free_tsk;
 327
 328         err = arch_dup_task_struct(tsk, orig);
 329         if (err)
 330                 goto free_ti;
 331
 332         tsk->stack = ti;
 333 #ifdef CONFIG_SECCOMP
 334         /*
 335          * We must handle setting up seccomp filters once we're under
 336          * the sighand lock in case orig has changed between now and
 337          * then. Until then, filter must be NULL to avoid messing up
 338          * the usage counts on the error path calling free_task.
 339          */
 340         tsk->seccomp.filter = NULL;
 341 #endif
 342
 343         setup_thread_stack(tsk, orig);
 344         clear_user_return_notifier(tsk);
 345         clear_tsk_need_resched(tsk);
 346         set_task_stack_end_magic(tsk);
 347
 348 #ifdef CONFIG_CC_STACKPROTECTOR
 349         tsk->stack_canary = get_random_int();
 350 #endif
 351
 352         /*
 353          * One for us, one for whoever does the "release_task()" (usually
 354          * parent)
 355          */
 356         atomic_set(&tsk->usage, 2);
 357 #ifdef CONFIG_BLK_DEV_IO_TRACE
 358         tsk->btrace_seq = 0;
 359 #endif
 360         tsk->splice_pipe = NULL;
 361         tsk->task_frag.page = NULL;
 362
 363         account_kernel_stack(ti, 1);
 364
 365         return tsk;
 366
 367 free_ti:
 368         free_thread_info(ti);
 369 free_tsk:
 370         free_task_struct(tsk);
 371         return NULL;
 372 }
 373
 374 #ifdef CONFIG_MMU
 375 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 376 {
 377         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 378         struct rb_node **rb_link, *rb_parent;
 379         int retval;
 380         unsigned long charge;
 381
 382         uprobe_start_dup_mmap();
 383         down_write(&oldmm->mmap_sem);
 384         flush_cache_dup_mm(oldmm);
 385         uprobe_dup_mmap(oldmm, mm);
 386         /*
 387          * Not linked in yet - no deadlock potential:
 388          */
 389         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 390
 391         mm->total_vm = oldmm->total_vm;
 392         mm->shared_vm = oldmm->shared_vm;
 393         mm->exec_vm = oldmm->exec_vm;
 394         mm->stack_vm = oldmm->stack_vm;
 395
 396         rb_link = &mm->mm_rb.rb_node;
 397         rb_parent = NULL;
 398         pprev = &mm->mmap;
 399         retval = ksm_fork(mm, oldmm);
 400         if (retval)
 401                 goto out;
 402         retval = khugepaged_fork(mm, oldmm);
 403         if (retval)
 404                 goto out;
 405
 406         prev = NULL;
 407         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 408                 struct file *file;
 409
 410                 if (mpnt->vm_flags & VM_DONTCOPY) {
 411                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 412                                                         -vma_pages(mpnt));
 413                         continue;
 414                 }
 415                 charge = 0;
 416                 if (mpnt->vm_flags & VM_ACCOUNT) {
 417                         unsigned long len = vma_pages(mpnt);
 418
 419                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 420                                 goto fail_nomem;
 421                         charge = len;
 422                 }
 423                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 424                 if (!tmp)
 425                         goto fail_nomem;
 426                 *tmp = *mpnt;
 427                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 428                 retval = vma_dup_policy(mpnt, tmp);
 429                 if (retval)
 430                         goto fail_nomem_policy;
 431                 tmp->vm_mm = mm;
 432                 if (anon_vma_fork(tmp, mpnt))
 433                         goto fail_nomem_anon_vma_fork;
 434                 tmp->vm_flags &= ~VM_LOCKED;
 435                 tmp->vm_next = tmp->vm_prev = NULL;
 436                 file = tmp->vm_file;
 437                 if (file) {
 438                         struct inode *inode = file_inode(file);
 439                         struct address_space *mapping = file->f_mapping;
 440
 441                         get_file(file);
 442                         if (tmp->vm_flags & VM_DENYWRITE)
 443                                 atomic_dec(&inode->i_writecount);
 444                         i_mmap_lock_write(mapping);
 445                         if (tmp->vm_flags & VM_SHARED)
 446                                 atomic_inc(&mapping->i_mmap_writable);
 447                         flush_dcache_mmap_lock(mapping);
 448                         /* insert tmp into the share list, just after mpnt */
 449                         vma_interval_tree_insert_after(tmp, mpnt,
 450                                         &mapping->i_mmap);
 451                         flush_dcache_mmap_unlock(mapping);
 452                         i_mmap_unlock_write(mapping);
 453                 }
 454
 455                 /*
 456                  * Clear hugetlb-related page reserves for children. This only
 457                  * affects MAP_PRIVATE mappings. Faults generated by the child
 458                  * are not guaranteed to succeed, even if read-only
 459                  */
 460                 if (is_vm_hugetlb_page(tmp))
 461                         reset_vma_resv_huge_pages(tmp);
 462
 463                 /*
 464                  * Link in the new vma and copy the page table entries.
 465                  */
 466                 *pprev = tmp;
 467                 pprev = &tmp->vm_next;
 468                 tmp->vm_prev = prev;
 469                 prev = tmp;
 470
 471                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 472                 rb_link = &tmp->vm_rb.rb_right;
 473                 rb_parent = &tmp->vm_rb;
 474
 475                 mm->map_count++;
 476                 retval = copy_page_range(mm, oldmm, mpnt);
 477
 478                 if (tmp->vm_ops && tmp->vm_ops->open)
 479                         tmp->vm_ops->open(tmp);
 480
 481                 if (retval)
 482                         goto out;
 483         }
 484         /* a new mm has just been created */
 485         arch_dup_mmap(oldmm, mm);
 486         retval = 0;
 487 out:
 488         up_write(&mm->mmap_sem);
 489         flush_tlb_mm(oldmm);
 490         up_write(&oldmm->mmap_sem);
 491         uprobe_end_dup_mmap();
 492         return retval;
 493 fail_nomem_anon_vma_fork:
 494         mpol_put(vma_policy(tmp));
 495 fail_nomem_policy:
 496         kmem_cache_free(vm_area_cachep, tmp);
 497 fail_nomem:
 498         retval = -ENOMEM;
 499         vm_unacct_memory(charge);
 500         goto out;
 501 }
 502
 503 static inline int mm_alloc_pgd(struct mm_struct *mm)
 504 {
 505         mm->pgd = pgd_alloc(mm);
 506         if (unlikely(!mm->pgd))
 507                 return -ENOMEM;
 508         return 0;
 509 }
 510
 511 static inline void mm_free_pgd(struct mm_struct *mm)
 512 {
 513         pgd_free(mm, mm->pgd);
 514 }
 515 #else
 516 #define dup_mmap(mm, oldmm)     (0)
 517 #define mm_alloc_pgd(mm)        (0)
 518 #define mm_free_pgd(mm)
 519 #endif /* CONFIG_MMU */
 520
 521 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 522
 523 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 524 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 525
 526 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 527
 528 static int __init coredump_filter_setup(char *s)
 529 {
 530         default_dump_filter =
 531                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 532                 MMF_DUMP_FILTER_MASK;
 533         return 1;
 534 }
 535
 536 __setup("coredump_filter=", coredump_filter_setup);
 537
 538 #include <linux/init_task.h>
 539
 540 static void mm_init_aio(struct mm_struct *mm)
 541 {
 542 #ifdef CONFIG_AIO
 543         spin_lock_init(&mm->ioctx_lock);
 544         mm->ioctx_table = NULL;
 545 #endif
 546 }
 547
 548 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 549 {
 550 #ifdef CONFIG_MEMCG
 551         mm->owner = p;
 552 #endif
 553 }
 554
 555 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 556 {
 557         mm->mmap = NULL;
 558         mm->mm_rb = RB_ROOT;
 559         mm->vmacache_seqnum = 0;
 560         atomic_set(&mm->mm_users, 1);
 561         atomic_set(&mm->mm_count, 1);
 562         init_rwsem(&mm->mmap_sem);
 563         INIT_LIST_HEAD(&mm->mmlist);
 564         mm->core_state = NULL;
 565         atomic_long_set(&mm->nr_ptes, 0);
 566         mm_nr_pmds_init(mm);
 567         mm->map_count = 0;
 568         mm->locked_vm = 0;
 569         mm->pinned_vm = 0;
 570         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 571         spin_lock_init(&mm->page_table_lock);
 572         mm_init_cpumask(mm);
 573         mm_init_aio(mm);
 574         mm_init_owner(mm, p);
 575         mmu_notifier_mm_init(mm);
 576         clear_tlb_flush_pending(mm);
 577 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 578         mm->pmd_huge_pte = NULL;
 579 #endif
 580
 581         if (current->mm) {
 582                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 583                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 584         } else {
 585                 mm->flags = default_dump_filter;
 586                 mm->def_flags = 0;
 587         }
 588
 589         if (mm_alloc_pgd(mm))
 590                 goto fail_nopgd;
 591
 592         if (init_new_context(p, mm))
 593                 goto fail_nocontext;
 594
 595         return mm;
 596
 597 fail_nocontext:
 598         mm_free_pgd(mm);
 599 fail_nopgd:
 600         free_mm(mm);
 601         return NULL;
 602 }
 603
 604 static void check_mm(struct mm_struct *mm)
 605 {
 606         int i;
 607
 608         for (i = 0; i < NR_MM_COUNTERS; i++) {
 609                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 610
 611                 if (unlikely(x))
 612                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 613                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 614         }
 615
 616         if (atomic_long_read(&mm->nr_ptes))
 617                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 618                                 atomic_long_read(&mm->nr_ptes));
 619         if (mm_nr_pmds(mm))
 620                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 621                                 mm_nr_pmds(mm));
 622
 623 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 624         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 625 #endif
 626 }
 627
 628 /*
 629  * Allocate and initialize an mm_struct.
 630  */
 631 struct mm_struct *mm_alloc(void)
 632 {
 633         struct mm_struct *mm;
 634
 635         mm = allocate_mm();
 636         if (!mm)
 637                 return NULL;
 638
 639         memset(mm, 0, sizeof(*mm));
 640         return mm_init(mm, current);
 641 }
 642
 643 /*
 644  * Called when the last reference to the mm
 645  * is dropped: either by a lazy thread or by
 646  * mmput. Free the page directory and the mm.
 647  */
 648 void __mmdrop(struct mm_struct *mm)
 649 {
 650         BUG_ON(mm == &init_mm);
 651         mm_free_pgd(mm);
 652         destroy_context(mm);
 653         mmu_notifier_mm_destroy(mm);
 654         check_mm(mm);
 655         free_mm(mm);
 656 }
 657 EXPORT_SYMBOL_GPL(__mmdrop);
 658
 659 /*
 660  * Decrement the use count and release all resources for an mm.
 661  */
 662 void mmput(struct mm_struct *mm)
 663 {
 664         might_sleep();
 665
 666         if (atomic_dec_and_test(&mm->mm_users)) {
 667                 uprobe_clear_state(mm);
 668                 exit_aio(mm);
 669                 ksm_exit(mm);
 670                 khugepaged_exit(mm); /* must run before exit_mmap */
 671                 exit_mmap(mm);
 672                 set_mm_exe_file(mm, NULL);
 673                 if (!list_empty(&mm->mmlist)) {
 674                         spin_lock(&mmlist_lock);
 675                         list_del(&mm->mmlist);
 676                         spin_unlock(&mmlist_lock);
 677                 }
 678                 if (mm->binfmt)
 679                         module_put(mm->binfmt->module);
 680                 mmdrop(mm);
 681         }
 682 }
 683 EXPORT_SYMBOL_GPL(mmput);
 684
 685 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 686 {
 687         if (new_exe_file)
 688                 get_file(new_exe_file);
 689         if (mm->exe_file)
 690                 fput(mm->exe_file);
 691         mm->exe_file = new_exe_file;
 692 }
 693
 694 struct file *get_mm_exe_file(struct mm_struct *mm)
 695 {
 696         struct file *exe_file;
 697
 698         /* We need mmap_sem to protect against races with removal of exe_file */
 699         down_read(&mm->mmap_sem);
 700         exe_file = mm->exe_file;
 701         if (exe_file)
 702                 get_file(exe_file);
 703         up_read(&mm->mmap_sem);
 704         return exe_file;
 705 }
 706
 707 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 708 {
 709         /* It's safe to write the exe_file pointer without exe_file_lock because
 710          * this is called during fork when the task is not yet in /proc */
 711         newmm->exe_file = get_mm_exe_file(oldmm);
 712 }
 713
 714 /**
 715  * get_task_mm - acquire a reference to the task's mm
 716  *
 717  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 718  * this kernel workthread has transiently adopted a user mm with use_mm,
 719  * to do its AIO) is not set and if so returns a reference to it, after
 720  * bumping up the use count.  User must release the mm via mmput()
 721  * after use.  Typically used by /proc and ptrace.
 722  */
 723 struct mm_struct *get_task_mm(struct task_struct *task)
 724 {
 725         struct mm_struct *mm;
 726
 727         task_lock(task);
 728         mm = task->mm;
 729         if (mm) {
 730                 if (task->flags & PF_KTHREAD)
 731                         mm = NULL;
 732                 else
 733                         atomic_inc(&mm->mm_users);
 734         }
 735         task_unlock(task);
 736         return mm;
 737 }
 738 EXPORT_SYMBOL_GPL(get_task_mm);
 739
 740 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 741 {
 742         struct mm_struct *mm;
 743         int err;
 744
 745         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 746         if (err)
 747                 return ERR_PTR(err);
 748
 749         mm = get_task_mm(task);
 750         if (mm && mm != current->mm &&
 751                         !ptrace_may_access(task, mode)) {
 752                 mmput(mm);
 753                 mm = ERR_PTR(-EACCES);
 754         }
 755         mutex_unlock(&task->signal->cred_guard_mutex);
 756
 757         return mm;
 758 }
 759
 760 static void complete_vfork_done(struct task_struct *tsk)
 761 {
 762         struct completion *vfork;
 763
 764         task_lock(tsk);
 765         vfork = tsk->vfork_done;
 766         if (likely(vfork)) {
 767                 tsk->vfork_done = NULL;
 768                 complete(vfork);
 769         }
 770         task_unlock(tsk);
 771 }
 772
 773 static int wait_for_vfork_done(struct task_struct *child,
 774                                 struct completion *vfork)
 775 {
 776         int killed;
 777
 778         freezer_do_not_count();
 779         killed = wait_for_completion_killable(vfork);
 780         freezer_count();
 781
 782         if (killed) {
 783                 task_lock(child);
 784                 child->vfork_done = NULL;
 785                 task_unlock(child);
 786         }
 787
 788         put_task_struct(child);
 789         return killed;
 790 }
 791
 792 /* Please note the differences between mmput and mm_release.
 793  * mmput is called whenever we stop holding onto a mm_struct,
 794  * error success whatever.
 795  *
 796  * mm_release is called after a mm_struct has been removed
 797  * from the current process.
 798  *
 799  * This difference is important for error handling, when we
 800  * only half set up a mm_struct for a new process and need to restore
 801  * the old one.  Because we mmput the new mm_struct before
 802  * restoring the old one. . .
 803  * Eric Biederman 10 January 1998
 804  */
 805 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 806 {
 807         /* Get rid of any futexes when releasing the mm */
 808 #ifdef CONFIG_FUTEX
 809         if (unlikely(tsk->robust_list)) {
 810                 exit_robust_list(tsk);
 811                 tsk->robust_list = NULL;
 812         }
 813 #ifdef CONFIG_COMPAT
 814         if (unlikely(tsk->compat_robust_list)) {
 815                 compat_exit_robust_list(tsk);
 816                 tsk->compat_robust_list = NULL;
 817         }
 818 #endif
 819         if (unlikely(!list_empty(&tsk->pi_state_list)))
 820                 exit_pi_state_list(tsk);
 821 #endif
 822
 823         uprobe_free_utask(tsk);
 824
 825         /* Get rid of any cached register state */
 826         deactivate_mm(tsk, mm);
 827
 828         /*
 829          * If we're exiting normally, clear a user-space tid field if
 830          * requested.  We leave this alone when dying by signal, to leave
 831          * the value intact in a core dump, and to save the unnecessary
 832          * trouble, say, a killed vfork parent shouldn't touch this mm.
 833          * Userland only wants this done for a sys_exit.
 834          */
 835         if (tsk->clear_child_tid) {
 836                 if (!(tsk->flags & PF_SIGNALED) &&
 837                     atomic_read(&mm->mm_users) > 1) {
 838                         /*
 839                          * We don't check the error code - if userspace has
 840                          * not set up a proper pointer then tough luck.
 841                          */
 842                         put_user(0, tsk->clear_child_tid);
 843                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 844                                         1, NULL, NULL, 0);
 845                 }
 846                 tsk->clear_child_tid = NULL;
 847         }
 848
 849         /*
 850          * All done, finally we can wake up parent and return this mm to him.
 851          * Also kthread_stop() uses this completion for synchronization.
 852          */
 853         if (tsk->vfork_done)
 854                 complete_vfork_done(tsk);
 855 }
 856
 857 /*
 858  * Allocate a new mm structure and copy contents from the
 859  * mm structure of the passed in task structure.
 860  */
 861 static struct mm_struct *dup_mm(struct task_struct *tsk)
 862 {
 863         struct mm_struct *mm, *oldmm = current->mm;
 864         int err;
 865
 866         mm = allocate_mm();
 867         if (!mm)
 868                 goto fail_nomem;
 869
 870         memcpy(mm, oldmm, sizeof(*mm));
 871
 872         if (!mm_init(mm, tsk))
 873                 goto fail_nomem;
 874
 875         dup_mm_exe_file(oldmm, mm);
 876
 877         err = dup_mmap(mm, oldmm);
 878         if (err)
 879                 goto free_pt;
 880
 881         mm->hiwater_rss = get_mm_rss(mm);
 882         mm->hiwater_vm = mm->total_vm;
 883
 884         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 885                 goto free_pt;
 886
 887         return mm;
 888
 889 free_pt:
 890         /* don't put binfmt in mmput, we haven't got module yet */
 891         mm->binfmt = NULL;
 892         mmput(mm);
 893
 894 fail_nomem:
 895         return NULL;
 896 }
 897
 898 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 899 {
 900         struct mm_struct *mm, *oldmm;
 901         int retval;
 902
 903         tsk->min_flt = tsk->maj_flt = 0;
 904         tsk->nvcsw = tsk->nivcsw = 0;
 905 #ifdef CONFIG_DETECT_HUNG_TASK
 906         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 907 #endif
 908
 909         tsk->mm = NULL;
 910         tsk->active_mm = NULL;
 911
 912         /*
 913          * Are we cloning a kernel thread?
 914          *
 915          * We need to steal a active VM for that..
 916          */
 917         oldmm = current->mm;
 918         if (!oldmm)
 919                 return 0;
 920
 921         /* initialize the new vmacache entries */
 922         vmacache_flush(tsk);
 923
 924         if (clone_flags & CLONE_VM) {
 925                 atomic_inc(&oldmm->mm_users);
 926                 mm = oldmm;
 927                 goto good_mm;
 928         }
 929
 930         retval = -ENOMEM;
 931         mm = dup_mm(tsk);
 932         if (!mm)
 933                 goto fail_nomem;
 934
 935 good_mm:
 936         tsk->mm = mm;
 937         tsk->active_mm = mm;
 938         return 0;
 939
 940 fail_nomem:
 941         return retval;
 942 }
 943
 944 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 945 {
 946         struct fs_struct *fs = current->fs;
 947         if (clone_flags & CLONE_FS) {
 948                 /* tsk->fs is already what we want */
 949                 spin_lock(&fs->lock);
 950                 if (fs->in_exec) {
 951                         spin_unlock(&fs->lock);
 952                         return -EAGAIN;
 953                 }
 954                 fs->users++;
 955                 spin_unlock(&fs->lock);
 956                 return 0;
 957         }
 958         tsk->fs = copy_fs_struct(fs);
 959         if (!tsk->fs)
 960                 return -ENOMEM;
 961         return 0;
 962 }
 963
 964 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 965 {
 966         struct files_struct *oldf, *newf;
 967         int error = 0;
 968
 969         /*
 970          * A background process may not have any files ...
 971          */
 972         oldf = current->files;
 973         if (!oldf)
 974                 goto out;
 975
 976         if (clone_flags & CLONE_FILES) {
 977                 atomic_inc(&oldf->count);
 978                 goto out;
 979         }
 980
 981         newf = dup_fd(oldf, &error);
 982         if (!newf)
 983                 goto out;
 984
 985         tsk->files = newf;
 986         error = 0;
 987 out:
 988         return error;
 989 }
 990
 991 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 992 {
 993 #ifdef CONFIG_BLOCK
 994         struct io_context *ioc = current->io_context;
 995         struct io_context *new_ioc;
 996
 997         if (!ioc)
 998                 return 0;
 999         /*
1000          * Share io context with parent, if CLONE_IO is set
1001          */
1002         if (clone_flags & CLONE_IO) {
1003                 ioc_task_link(ioc);
1004                 tsk->io_context = ioc;
1005         } else if (ioprio_valid(ioc->ioprio)) {
1006                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1007                 if (unlikely(!new_ioc))
1008                         return -ENOMEM;
1009
1010                 new_ioc->ioprio = ioc->ioprio;
1011                 put_io_context(new_ioc);
1012         }
1013 #endif
1014         return 0;
1015 }
1016
1017 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1018 {
1019         struct sighand_struct *sig;
1020
1021         if (clone_flags & CLONE_SIGHAND) {
1022                 atomic_inc(&current->sighand->count);
1023                 return 0;
1024         }
1025         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1026         rcu_assign_pointer(tsk->sighand, sig);
1027         if (!sig)
1028                 return -ENOMEM;
1029         atomic_set(&sig->count, 1);
1030         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1031         return 0;
1032 }
1033
1034 void __cleanup_sighand(struct sighand_struct *sighand)
1035 {
1036         if (atomic_dec_and_test(&sighand->count)) {
1037                 signalfd_cleanup(sighand);
1038                 /*
1039                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1040                  * without an RCU grace period, see __lock_task_sighand().
1041                  */
1042                 kmem_cache_free(sighand_cachep, sighand);
1043         }
1044 }
1045
1046 /*
1047  * Initialize POSIX timer handling for a thread group.
1048  */
1049 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1050 {
1051         unsigned long cpu_limit;
1052
1053         /* Thread group counters. */
1054         thread_group_cputime_init(sig);
1055
1056         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1057         if (cpu_limit != RLIM_INFINITY) {
1058                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1059                 sig->cputimer.running = 1;
1060         }
1061
1062         /* The timer lists. */
1063         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1064         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1065         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1066 }
1067
1068 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1069 {
1070         struct signal_struct *sig;
1071
1072         if (clone_flags & CLONE_THREAD)
1073                 return 0;
1074
1075         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1076         tsk->signal = sig;
1077         if (!sig)
1078                 return -ENOMEM;
1079
1080         sig->nr_threads = 1;
1081         atomic_set(&sig->live, 1);
1082         atomic_set(&sig->sigcnt, 1);
1083
1084         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1085         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1086         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1087
1088         init_waitqueue_head(&sig->wait_chldexit);
1089         sig->curr_target = tsk;
1090         init_sigpending(&sig->shared_pending);
1091         INIT_LIST_HEAD(&sig->posix_timers);
1092         seqlock_init(&sig->stats_lock);
1093
1094         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1095         sig->real_timer.function = it_real_fn;
1096
1097         task_lock(current->group_leader);
1098         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1099         task_unlock(current->group_leader);
1100
1101         posix_cpu_timers_init_group(sig);
1102
1103         tty_audit_fork(sig);
1104         sched_autogroup_fork(sig);
1105
1106 #ifdef CONFIG_CGROUPS
1107         init_rwsem(&sig->group_rwsem);
1108 #endif
1109
1110         sig->oom_score_adj = current->signal->oom_score_adj;
1111         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1112
1113         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1114                                    current->signal->is_child_subreaper;
1115
1116         mutex_init(&sig->cred_guard_mutex);
1117
1118         return 0;
1119 }
1120
1121 static void copy_seccomp(struct task_struct *p)
1122 {
1123 #ifdef CONFIG_SECCOMP
1124         /*
1125          * Must be called with sighand->lock held, which is common to
1126          * all threads in the group. Holding cred_guard_mutex is not
1127          * needed because this new task is not yet running and cannot
1128          * be racing exec.
1129          */
1130         assert_spin_locked(&current->sighand->siglock);
1131
1132         /* Ref-count the new filter user, and assign it. */
1133         get_seccomp_filter(current);
1134         p->seccomp = current->seccomp;
1135
1136         /*
1137          * Explicitly enable no_new_privs here in case it got set
1138          * between the task_struct being duplicated and holding the
1139          * sighand lock. The seccomp state and nnp must be in sync.
1140          */
1141         if (task_no_new_privs(current))
1142                 task_set_no_new_privs(p);
1143
1144         /*
1145          * If the parent gained a seccomp mode after copying thread
1146          * flags and between before we held the sighand lock, we have
1147          * to manually enable the seccomp thread flag here.
1148          */
1149         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1150                 set_tsk_thread_flag(p, TIF_SECCOMP);
1151 #endif
1152 }
1153
1154 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1155 {
1156         current->clear_child_tid = tidptr;
1157
1158         return task_pid_vnr(current);
1159 }
1160
1161 static void rt_mutex_init_task(struct task_struct *p)
1162 {
1163         raw_spin_lock_init(&p->pi_lock);
1164 #ifdef CONFIG_RT_MUTEXES
1165         p->pi_waiters = RB_ROOT;
1166         p->pi_waiters_leftmost = NULL;
1167         p->pi_blocked_on = NULL;
1168 #endif
1169 }
1170
1171 /*
1172  * Initialize POSIX timer handling for a single task.
1173  */
1174 static void posix_cpu_timers_init(struct task_struct *tsk)
1175 {
1176         tsk->cputime_expires.prof_exp = 0;
1177         tsk->cputime_expires.virt_exp = 0;
1178         tsk->cputime_expires.sched_exp = 0;
1179         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1180         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1181         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1182 }
1183
1184 static inline void
1185 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1186 {
1187          task->pids[type].pid = pid;
1188 }
1189
1190 /*
1191  * This creates a new process as a copy of the old one,
1192  * but does not actually start it yet.
1193  *
1194  * It copies the registers, and all the appropriate
1195  * parts of the process environment (as per the clone
1196  * flags). The actual kick-off is left to the caller.
1197  */
1198 static struct task_struct *copy_process(unsigned long clone_flags,
1199                                         unsigned long stack_start,
1200                                         unsigned long stack_size,
1201                                         int __user *child_tidptr,
1202                                         struct pid *pid,
1203                                         int trace)
1204 {
1205         int retval;
1206         struct task_struct *p;
1207
1208         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1209                 return ERR_PTR(-EINVAL);
1210
1211         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1212                 return ERR_PTR(-EINVAL);
1213
1214         /*
1215          * Thread groups must share signals as well, and detached threads
1216          * can only be started up within the thread group.
1217          */
1218         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1219                 return ERR_PTR(-EINVAL);
1220
1221         /*
1222          * Shared signal handlers imply shared VM. By way of the above,
1223          * thread groups also imply shared VM. Blocking this case allows
1224          * for various simplifications in other code.
1225          */
1226         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1227                 return ERR_PTR(-EINVAL);
1228
1229         /*
1230          * Siblings of global init remain as zombies on exit since they are
1231          * not reaped by their parent (swapper). To solve this and to avoid
1232          * multi-rooted process trees, prevent global and container-inits
1233          * from creating siblings.
1234          */
1235         if ((clone_flags & CLONE_PARENT) &&
1236                                 current->signal->flags & SIGNAL_UNKILLABLE)
1237                 return ERR_PTR(-EINVAL);
1238
1239         /*
1240          * If the new process will be in a different pid or user namespace
1241          * do not allow it to share a thread group or signal handlers or
1242          * parent with the forking task.
1243          */
1244         if (clone_flags & CLONE_SIGHAND) {
1245                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1246                     (task_active_pid_ns(current) !=
1247                                 current->nsproxy->pid_ns_for_children))
1248                         return ERR_PTR(-EINVAL);
1249         }
1250
1251         retval = security_task_create(clone_flags);
1252         if (retval)
1253                 goto fork_out;
1254
1255         retval = -ENOMEM;
1256         p = dup_task_struct(current);
1257         if (!p)
1258                 goto fork_out;
1259
1260         ftrace_graph_init_task(p);
1261
1262         rt_mutex_init_task(p);
1263
1264 #ifdef CONFIG_PROVE_LOCKING
1265         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1266         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1267 #endif
1268         retval = -EAGAIN;
1269         if (atomic_read(&p->real_cred->user->processes) >=
1270                         task_rlimit(p, RLIMIT_NPROC)) {
1271                 if (p->real_cred->user != INIT_USER &&
1272                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1273                         goto bad_fork_free;
1274         }
1275         current->flags &= ~PF_NPROC_EXCEEDED;
1276
1277         retval = copy_creds(p, clone_flags);
1278         if (retval < 0)
1279                 goto bad_fork_free;
1280
1281         /*
1282          * If multiple threads are within copy_process(), then this check
1283          * triggers too late. This doesn't hurt, the check is only there
1284          * to stop root fork bombs.
1285          */
1286         retval = -EAGAIN;
1287         if (nr_threads >= max_threads)
1288                 goto bad_fork_cleanup_count;
1289
1290         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1291         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1292         p->flags |= PF_FORKNOEXEC;
1293         INIT_LIST_HEAD(&p->children);
1294         INIT_LIST_HEAD(&p->sibling);
1295         rcu_copy_process(p);
1296         p->vfork_done = NULL;
1297         spin_lock_init(&p->alloc_lock);
1298
1299         init_sigpending(&p->pending);
1300
1301         p->utime = p->stime = p->gtime = 0;
1302         p->utimescaled = p->stimescaled = 0;
1303 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1304         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1305 #endif
1306 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1307         seqlock_init(&p->vtime_seqlock);
1308         p->vtime_snap = 0;
1309         p->vtime_snap_whence = VTIME_SLEEPING;
1310 #endif
1311
1312 #if defined(SPLIT_RSS_COUNTING)
1313         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1314 #endif
1315
1316         p->default_timer_slack_ns = current->timer_slack_ns;
1317
1318         task_io_accounting_init(&p->ioac);
1319         acct_clear_integrals(p);
1320
1321         posix_cpu_timers_init(p);
1322
1323         p->start_time = ktime_get_ns();
1324         p->real_start_time = ktime_get_boot_ns();
1325         p->io_context = NULL;
1326         p->audit_context = NULL;
1327         if (clone_flags & CLONE_THREAD)
1328                 threadgroup_change_begin(current);
1329         cgroup_fork(p);
1330 #ifdef CONFIG_NUMA
1331         p->mempolicy = mpol_dup(p->mempolicy);
1332         if (IS_ERR(p->mempolicy)) {
1333                 retval = PTR_ERR(p->mempolicy);
1334                 p->mempolicy = NULL;
1335                 goto bad_fork_cleanup_threadgroup_lock;
1336         }
1337 #endif
1338 #ifdef CONFIG_CPUSETS
1339         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1340         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1341         seqcount_init(&p->mems_allowed_seq);
1342 #endif
1343 #ifdef CONFIG_TRACE_IRQFLAGS
1344         p->irq_events = 0;
1345         p->hardirqs_enabled = 0;
1346         p->hardirq_enable_ip = 0;
1347         p->hardirq_enable_event = 0;
1348         p->hardirq_disable_ip = _THIS_IP_;
1349         p->hardirq_disable_event = 0;
1350         p->softirqs_enabled = 1;
1351         p->softirq_enable_ip = _THIS_IP_;
1352         p->softirq_enable_event = 0;
1353         p->softirq_disable_ip = 0;
1354         p->softirq_disable_event = 0;
1355         p->hardirq_context = 0;
1356         p->softirq_context = 0;
1357 #endif
1358 #ifdef CONFIG_LOCKDEP
1359         p->lockdep_depth = 0; /* no locks held yet */
1360         p->curr_chain_key = 0;
1361         p->lockdep_recursion = 0;
1362 #endif
1363
1364 #ifdef CONFIG_DEBUG_MUTEXES
1365         p->blocked_on = NULL; /* not blocked yet */
1366 #endif
1367 #ifdef CONFIG_BCACHE
1368         p->sequential_io        = 0;
1369         p->sequential_io_avg    = 0;
1370 #endif
1371
1372         /* Perform scheduler related setup. Assign this task to a CPU. */
1373         retval = sched_fork(clone_flags, p);
1374         if (retval)
1375                 goto bad_fork_cleanup_policy;
1376
1377         retval = perf_event_init_task(p);
1378         if (retval)
1379                 goto bad_fork_cleanup_policy;
1380         retval = audit_alloc(p);
1381         if (retval)
1382                 goto bad_fork_cleanup_perf;
1383         /* copy all the process information */
1384         shm_init_task(p);
1385         retval = copy_semundo(clone_flags, p);
1386         if (retval)
1387                 goto bad_fork_cleanup_audit;
1388         retval = copy_files(clone_flags, p);
1389         if (retval)
1390                 goto bad_fork_cleanup_semundo;
1391         retval = copy_fs(clone_flags, p);
1392         if (retval)
1393                 goto bad_fork_cleanup_files;
1394         retval = copy_sighand(clone_flags, p);
1395         if (retval)
1396                 goto bad_fork_cleanup_fs;
1397         retval = copy_signal(clone_flags, p);
1398         if (retval)
1399                 goto bad_fork_cleanup_sighand;
1400         retval = copy_mm(clone_flags, p);
1401         if (retval)
1402                 goto bad_fork_cleanup_signal;
1403         retval = copy_namespaces(clone_flags, p);
1404         if (retval)
1405                 goto bad_fork_cleanup_mm;
1406         retval = copy_io(clone_flags, p);
1407         if (retval)
1408                 goto bad_fork_cleanup_namespaces;
1409         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1410         if (retval)
1411                 goto bad_fork_cleanup_io;
1412
1413         if (pid != &init_struct_pid) {
1414                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1415                 if (IS_ERR(pid)) {
1416                         retval = PTR_ERR(pid);
1417                         goto bad_fork_cleanup_io;
1418                 }
1419         }
1420
1421         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1422         /*
1423          * Clear TID on mm_release()?
1424          */
1425         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1426 #ifdef CONFIG_BLOCK
1427         p->plug = NULL;
1428 #endif
1429 #ifdef CONFIG_FUTEX
1430         p->robust_list = NULL;
1431 #ifdef CONFIG_COMPAT
1432         p->compat_robust_list = NULL;
1433 #endif
1434         INIT_LIST_HEAD(&p->pi_state_list);
1435         p->pi_state_cache = NULL;
1436 #endif
1437         /*
1438          * sigaltstack should be cleared when sharing the same VM
1439          */
1440         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1441                 p->sas_ss_sp = p->sas_ss_size = 0;
1442
1443         /*
1444          * Syscall tracing and stepping should be turned off in the
1445          * child regardless of CLONE_PTRACE.
1446          */
1447         user_disable_single_step(p);
1448         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1449 #ifdef TIF_SYSCALL_EMU
1450         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1451 #endif
1452         clear_all_latency_tracing(p);
1453
1454         /* ok, now we should be set up.. */
1455         p->pid = pid_nr(pid);
1456         if (clone_flags & CLONE_THREAD) {
1457                 p->exit_signal = -1;
1458                 p->group_leader = current->group_leader;
1459                 p->tgid = current->tgid;
1460         } else {
1461                 if (clone_flags & CLONE_PARENT)
1462                         p->exit_signal = current->group_leader->exit_signal;
1463                 else
1464                         p->exit_signal = (clone_flags & CSIGNAL);
1465                 p->group_leader = p;
1466                 p->tgid = p->pid;
1467         }
1468
1469         p->nr_dirtied = 0;
1470         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1471         p->dirty_paused_when = 0;
1472
1473         p->pdeath_signal = 0;
1474         INIT_LIST_HEAD(&p->thread_group);
1475         p->task_works = NULL;
1476
1477         /*
1478          * Make it visible to the rest of the system, but dont wake it up yet.
1479          * Need tasklist lock for parent etc handling!
1480          */
1481         write_lock_irq(&tasklist_lock);
1482
1483         /* CLONE_PARENT re-uses the old parent */
1484         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1485                 p->real_parent = current->real_parent;
1486                 p->parent_exec_id = current->parent_exec_id;
1487         } else {
1488                 p->real_parent = current;
1489                 p->parent_exec_id = current->self_exec_id;
1490         }
1491
1492         spin_lock(&current->sighand->siglock);
1493
1494         /*
1495          * Copy seccomp details explicitly here, in case they were changed
1496          * before holding sighand lock.
1497          */
1498         copy_seccomp(p);
1499
1500         /*
1501          * Process group and session signals need to be delivered to just the
1502          * parent before the fork or both the parent and the child after the
1503          * fork. Restart if a signal comes in before we add the new process to
1504          * it's process group.
1505          * A fatal signal pending means that current will exit, so the new
1506          * thread can't slip out of an OOM kill (or normal SIGKILL).
1507         */
1508         recalc_sigpending();
1509         if (signal_pending(current)) {
1510                 spin_unlock(&current->sighand->siglock);
1511                 write_unlock_irq(&tasklist_lock);
1512                 retval = -ERESTARTNOINTR;
1513                 goto bad_fork_free_pid;
1514         }
1515
1516         if (likely(p->pid)) {
1517                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1518
1519                 init_task_pid(p, PIDTYPE_PID, pid);
1520                 if (thread_group_leader(p)) {
1521                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1522                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1523
1524                         if (is_child_reaper(pid)) {
1525                                 ns_of_pid(pid)->child_reaper = p;
1526                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1527                         }
1528
1529                         p->signal->leader_pid = pid;
1530                         p->signal->tty = tty_kref_get(current->signal->tty);
1531                         list_add_tail(&p->sibling, &p->real_parent->children);
1532                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1533                         attach_pid(p, PIDTYPE_PGID);
1534                         attach_pid(p, PIDTYPE_SID);
1535                         __this_cpu_inc(process_counts);
1536                 } else {
1537                         current->signal->nr_threads++;
1538                         atomic_inc(&current->signal->live);
1539                         atomic_inc(&current->signal->sigcnt);
1540                         list_add_tail_rcu(&p->thread_group,
1541                                           &p->group_leader->thread_group);
1542                         list_add_tail_rcu(&p->thread_node,
1543                                           &p->signal->thread_head);
1544                 }
1545                 attach_pid(p, PIDTYPE_PID);
1546                 nr_threads++;
1547         }
1548
1549         total_forks++;
1550         spin_unlock(&current->sighand->siglock);
1551         syscall_tracepoint_update(p);
1552         write_unlock_irq(&tasklist_lock);
1553
1554         proc_fork_connector(p);
1555         cgroup_post_fork(p);
1556         if (clone_flags & CLONE_THREAD)
1557                 threadgroup_change_end(current);
1558         perf_event_fork(p);
1559
1560         trace_task_newtask(p, clone_flags);
1561         uprobe_copy_process(p, clone_flags);
1562
1563         return p;
1564
1565 bad_fork_free_pid:
1566         if (pid != &init_struct_pid)
1567                 free_pid(pid);
1568 bad_fork_cleanup_io:
1569         if (p->io_context)
1570                 exit_io_context(p);
1571 bad_fork_cleanup_namespaces:
1572         exit_task_namespaces(p);
1573 bad_fork_cleanup_mm:
1574         if (p->mm)
1575                 mmput(p->mm);
1576 bad_fork_cleanup_signal:
1577         if (!(clone_flags & CLONE_THREAD))
1578                 free_signal_struct(p->signal);
1579 bad_fork_cleanup_sighand:
1580         __cleanup_sighand(p->sighand);
1581 bad_fork_cleanup_fs:
1582         exit_fs(p); /* blocking */
1583 bad_fork_cleanup_files:
1584         exit_files(p); /* blocking */
1585 bad_fork_cleanup_semundo:
1586         exit_sem(p);
1587 bad_fork_cleanup_audit:
1588         audit_free(p);
1589 bad_fork_cleanup_perf:
1590         perf_event_free_task(p);
1591 bad_fork_cleanup_policy:
1592 #ifdef CONFIG_NUMA
1593         mpol_put(p->mempolicy);
1594 bad_fork_cleanup_threadgroup_lock:
1595 #endif
1596         if (clone_flags & CLONE_THREAD)
1597                 threadgroup_change_end(current);
1598         delayacct_tsk_free(p);
1599 bad_fork_cleanup_count:
1600         atomic_dec(&p->cred->user->processes);
1601         exit_creds(p);
1602 bad_fork_free:
1603         free_task(p);
1604 fork_out:
1605         return ERR_PTR(retval);
1606 }
1607
1608 static inline void init_idle_pids(struct pid_link *links)
1609 {
1610         enum pid_type type;
1611
1612         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1613                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1614                 links[type].pid = &init_struct_pid;
1615         }
1616 }
1617
1618 struct task_struct *fork_idle(int cpu)
1619 {
1620         struct task_struct *task;
1621         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1622         if (!IS_ERR(task)) {
1623                 init_idle_pids(task->pids);
1624                 init_idle(task, cpu);
1625         }
1626
1627         return task;
1628 }
1629
1630 /*
1631  *  Ok, this is the main fork-routine.
1632  *
1633  * It copies the process, and if successful kick-starts
1634  * it and waits for it to finish using the VM if required.
1635  */
1636 long do_fork(unsigned long clone_flags,
1637               unsigned long stack_start,
1638               unsigned long stack_size,
1639               int __user *parent_tidptr,
1640               int __user *child_tidptr)
1641 {
1642         struct task_struct *p;
1643         int trace = 0;
1644         long nr;
1645
1646         /*
1647          * Determine whether and which event to report to ptracer.  When
1648          * called from kernel_thread or CLONE_UNTRACED is explicitly
1649          * requested, no event is reported; otherwise, report if the event
1650          * for the type of forking is enabled.
1651          */
1652         if (!(clone_flags & CLONE_UNTRACED)) {
1653                 if (clone_flags & CLONE_VFORK)
1654                         trace = PTRACE_EVENT_VFORK;
1655                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1656                         trace = PTRACE_EVENT_CLONE;
1657                 else
1658                         trace = PTRACE_EVENT_FORK;
1659
1660                 if (likely(!ptrace_event_enabled(current, trace)))
1661                         trace = 0;
1662         }
1663
1664         p = copy_process(clone_flags, stack_start, stack_size,
1665                          child_tidptr, NULL, trace);
1666         /*
1667          * Do this prior waking up the new thread - the thread pointer
1668          * might get invalid after that point, if the thread exits quickly.
1669          */
1670         if (!IS_ERR(p)) {
1671                 struct completion vfork;
1672                 struct pid *pid;
1673
1674                 trace_sched_process_fork(current, p);
1675
1676                 pid = get_task_pid(p, PIDTYPE_PID);
1677                 nr = pid_vnr(pid);
1678
1679                 if (clone_flags & CLONE_PARENT_SETTID)
1680                         put_user(nr, parent_tidptr);
1681
1682                 if (clone_flags & CLONE_VFORK) {
1683                         p->vfork_done = &vfork;
1684                         init_completion(&vfork);
1685                         get_task_struct(p);
1686                 }
1687
1688                 wake_up_new_task(p);
1689
1690                 /* forking complete and child started to run, tell ptracer */
1691                 if (unlikely(trace))
1692                         ptrace_event_pid(trace, pid);
1693
1694                 if (clone_flags & CLONE_VFORK) {
1695                         if (!wait_for_vfork_done(p, &vfork))
1696                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1697                 }
1698
1699                 put_pid(pid);
1700         } else {
1701                 nr = PTR_ERR(p);
1702         }
1703         return nr;
1704 }
1705
1706 /*
1707  * Create a kernel thread.
1708  */
1709 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1710 {
1711         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1712                 (unsigned long)arg, NULL, NULL);
1713 }
1714
1715 #ifdef __ARCH_WANT_SYS_FORK
1716 SYSCALL_DEFINE0(fork)
1717 {
1718 #ifdef CONFIG_MMU
1719         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1720 #else
1721         /* can not support in nommu mode */
1722         return -EINVAL;
1723 #endif
1724 }
1725 #endif
1726
1727 #ifdef __ARCH_WANT_SYS_VFORK
1728 SYSCALL_DEFINE0(vfork)
1729 {
1730         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1731                         0, NULL, NULL);
1732 }
1733 #endif
1734
1735 #ifdef __ARCH_WANT_SYS_CLONE
1736 #ifdef CONFIG_CLONE_BACKWARDS
1737 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1738                  int __user *, parent_tidptr,
1739                  int, tls_val,
1740                  int __user *, child_tidptr)
1741 #elif defined(CONFIG_CLONE_BACKWARDS2)
1742 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1743                  int __user *, parent_tidptr,
1744                  int __user *, child_tidptr,
1745                  int, tls_val)
1746 #elif defined(CONFIG_CLONE_BACKWARDS3)
1747 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1748                 int, stack_size,
1749                 int __user *, parent_tidptr,
1750                 int __user *, child_tidptr,
1751                 int, tls_val)
1752 #else
1753 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1754                  int __user *, parent_tidptr,
1755                  int __user *, child_tidptr,
1756                  int, tls_val)
1757 #endif
1758 {
1759         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1760 }
1761 #endif
1762
1763 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1764 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1765 #endif
1766
1767 static void sighand_ctor(void *data)
1768 {
1769         struct sighand_struct *sighand = data;
1770
1771         spin_lock_init(&sighand->siglock);
1772         init_waitqueue_head(&sighand->signalfd_wqh);
1773 }
1774
1775 void __init proc_caches_init(void)
1776 {
1777         sighand_cachep = kmem_cache_create("sighand_cache",
1778                         sizeof(struct sighand_struct), 0,
1779                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1780                         SLAB_NOTRACK, sighand_ctor);
1781         signal_cachep = kmem_cache_create("signal_cache",
1782                         sizeof(struct signal_struct), 0,
1783                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1784         files_cachep = kmem_cache_create("files_cache",
1785                         sizeof(struct files_struct), 0,
1786                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1787         fs_cachep = kmem_cache_create("fs_cache",
1788                         sizeof(struct fs_struct), 0,
1789                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1790         /*
1791          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1792          * whole struct cpumask for the OFFSTACK case. We could change
1793          * this to *only* allocate as much of it as required by the
1794          * maximum number of CPU's we can ever have.  The cpumask_allocation
1795          * is at the end of the structure, exactly for that reason.
1796          */
1797         mm_cachep = kmem_cache_create("mm_struct",
1798                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1799                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1800         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1801         mmap_init();
1802         nsproxy_cache_init();
1803 }
1804
1805 /*
1806  * Check constraints on flags passed to the unshare system call.
1807  */
1808 static int check_unshare_flags(unsigned long unshare_flags)
1809 {
1810         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1811                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1812                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1813                                 CLONE_NEWUSER|CLONE_NEWPID))
1814                 return -EINVAL;
1815         /*
1816          * Not implemented, but pretend it works if there is nothing to
1817          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1818          * needs to unshare vm.
1819          */
1820         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1821                 /* FIXME: get_task_mm() increments ->mm_users */
1822                 if (atomic_read(&current->mm->mm_users) > 1)
1823                         return -EINVAL;
1824         }
1825
1826         return 0;
1827 }
1828
1829 /*
1830  * Unshare the filesystem structure if it is being shared
1831  */
1832 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1833 {
1834         struct fs_struct *fs = current->fs;
1835
1836         if (!(unshare_flags & CLONE_FS) || !fs)
1837                 return 0;
1838
1839         /* don't need lock here; in the worst case we'll do useless copy */
1840         if (fs->users == 1)
1841                 return 0;
1842
1843         *new_fsp = copy_fs_struct(fs);
1844         if (!*new_fsp)
1845                 return -ENOMEM;
1846
1847         return 0;
1848 }
1849
1850 /*
1851  * Unshare file descriptor table if it is being shared
1852  */
1853 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1854 {
1855         struct files_struct *fd = current->files;
1856         int error = 0;
1857
1858         if ((unshare_flags & CLONE_FILES) &&
1859             (fd && atomic_read(&fd->count) > 1)) {
1860                 *new_fdp = dup_fd(fd, &error);
1861                 if (!*new_fdp)
1862                         return error;
1863         }
1864
1865         return 0;
1866 }
1867
1868 /*
1869  * unshare allows a process to 'unshare' part of the process
1870  * context which was originally shared using clone.  copy_*
1871  * functions used by do_fork() cannot be used here directly
1872  * because they modify an inactive task_struct that is being
1873  * constructed. Here we are modifying the current, active,
1874  * task_struct.
1875  */
1876 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1877 {
1878         struct fs_struct *fs, *new_fs = NULL;
1879         struct files_struct *fd, *new_fd = NULL;
1880         struct cred *new_cred = NULL;
1881         struct nsproxy *new_nsproxy = NULL;
1882         int do_sysvsem = 0;
1883         int err;
1884
1885         /*
1886          * If unsharing a user namespace must also unshare the thread.
1887          */
1888         if (unshare_flags & CLONE_NEWUSER)
1889                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1890         /*
1891          * If unsharing a thread from a thread group, must also unshare vm.
1892          */
1893         if (unshare_flags & CLONE_THREAD)
1894                 unshare_flags |= CLONE_VM;
1895         /*
1896          * If unsharing vm, must also unshare signal handlers.
1897          */
1898         if (unshare_flags & CLONE_VM)
1899                 unshare_flags |= CLONE_SIGHAND;
1900         /*
1901          * If unsharing namespace, must also unshare filesystem information.
1902          */
1903         if (unshare_flags & CLONE_NEWNS)
1904                 unshare_flags |= CLONE_FS;
1905
1906         err = check_unshare_flags(unshare_flags);
1907         if (err)
1908                 goto bad_unshare_out;
1909         /*
1910          * CLONE_NEWIPC must also detach from the undolist: after switching
1911          * to a new ipc namespace, the semaphore arrays from the old
1912          * namespace are unreachable.
1913          */
1914         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1915                 do_sysvsem = 1;
1916         err = unshare_fs(unshare_flags, &new_fs);
1917         if (err)
1918                 goto bad_unshare_out;
1919         err = unshare_fd(unshare_flags, &new_fd);
1920         if (err)
1921                 goto bad_unshare_cleanup_fs;
1922         err = unshare_userns(unshare_flags, &new_cred);
1923         if (err)
1924                 goto bad_unshare_cleanup_fd;
1925         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1926                                          new_cred, new_fs);
1927         if (err)
1928                 goto bad_unshare_cleanup_cred;
1929
1930         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1931                 if (do_sysvsem) {
1932                         /*
1933                          * CLONE_SYSVSEM is equivalent to sys_exit().
1934                          */
1935                         exit_sem(current);
1936                 }
1937                 if (unshare_flags & CLONE_NEWIPC) {
1938                         /* Orphan segments in old ns (see sem above). */
1939                         exit_shm(current);
1940                         shm_init_task(current);
1941                 }
1942
1943                 if (new_nsproxy)
1944                         switch_task_namespaces(current, new_nsproxy);
1945
1946                 task_lock(current);
1947
1948                 if (new_fs) {
1949                         fs = current->fs;
1950                         spin_lock(&fs->lock);
1951                         current->fs = new_fs;
1952                         if (--fs->users)
1953                                 new_fs = NULL;
1954                         else
1955                                 new_fs = fs;
1956                         spin_unlock(&fs->lock);
1957                 }
1958
1959                 if (new_fd) {
1960                         fd = current->files;
1961                         current->files = new_fd;
1962                         new_fd = fd;
1963                 }
1964
1965                 task_unlock(current);
1966
1967                 if (new_cred) {
1968                         /* Install the new user namespace */
1969                         commit_creds(new_cred);
1970                         new_cred = NULL;
1971                 }
1972         }
1973
1974 bad_unshare_cleanup_cred:
1975         if (new_cred)
1976                 put_cred(new_cred);
1977 bad_unshare_cleanup_fd:
1978         if (new_fd)
1979                 put_files_struct(new_fd);
1980
1981 bad_unshare_cleanup_fs:
1982         if (new_fs)
1983                 free_fs_struct(new_fs);
1984
1985 bad_unshare_out:
1986         return err;
1987 }
1988
1989 /*
1990  *      Helper to unshare the files of the current task.
1991  *      We don't want to expose copy_files internals to
1992  *      the exec layer of the kernel.
1993  */
1994
1995 int unshare_files(struct files_struct **displaced)
1996 {
1997         struct task_struct *task = current;
1998         struct files_struct *copy = NULL;
1999         int error;
2000
2001         error = unshare_fd(CLONE_FILES, &copy);
2002         if (error || !copy) {
2003                 *displaced = NULL;
2004                 return error;
2005         }
2006         *displaced = task->files;
2007         task_lock(task);
2008         task->files = copy;
2009         task_unlock(task);
2010         return 0;
2011 }